Skip to main content
Internet Archive's 25th Anniversary Logo

Full text of "Analysis of dc offset in iOS devices for use in audio forensic examination"

See other formats


ANALYSIS OF DC OFFSET IN iOS DEVICES 
FOR USE IN AUDIO FORENSIC EXAMINATIONS 



by 



Daniel Bradley Fuller 
B.S., Middle Tennessee State University, 2006 



A thesis submitted to the 

Faculty of the Graduate School of the 

University of Colorado Denver in partial fulfillment 

of the requirements for the degree of 

Master of Science 

Media Forensics 

2013 



© 2013 by Daniel Bradley Fuller 
All Rights Reserved 



This thesis for the Master of Science degree by 

Daniel Bradley Fuller 

has been approved for the 

Department of Music and Entertainment Studies 

by 



Catalin Grigoras, Chair 



Jeff M. Smith 



Marcus K. Rogers 



May 4, 2013 



Fuller, Daniel Bradley (M.S. Media Forensics) 

Analysis of DC Offset for iOS Devices for Use in Audio Forensic Examinations 

Thesis directed by Professor Catalin Grigoras 

ABSTRACT 

Due to the physical properties of electronic components, DC offset will occur to 
some extent in all audio recordings. DC offset is the effect of direct current on an 
audio recording, and causes the audio signal to no longer oscillate around the 
absolute zero quantization level. The mean amplitude of a recording is calculated to 
determine the global amount of DC offset. Measuring the offset, its change over 
time, and its standard deviation can be used during forensic examination to aid in 
determining the authenticity of a recording as well as for exclusionary purposes 
when multiple recorders could possibly be the source of a recording. The scope of 
this thesis is to measure the DC offset that occurs in recordings made by Apple 
mobile devices running on iOS, to quantify the uniqueness of this offset within this 
family of devices as well as against previously tested audio recording devices, and to 
see if different hardware and apps affect the offset. To accurately determine this, 
multiple apps were tested in conjunction with the built-in microphone, the Apple 
EarPods that come with the iPhone 5, and the Apple Earphones that come with 
previous iPhone models. Furthermore, all recordings were made in laboratory 
conditions with a minimum amount of outside noise, only the app making the 
current recording was open, Auto-Brightness was switched off, and all outside 
connectivity (wireless, Bluetooth, 3G, 4G, LTE) was turned off. 

The form and content of this abstract are approved. I recommend its publication. 

Approved: Catalin Grigoras 



in 



DEDICATION 



I dedicate this thesis to my family, Tom, Josie and Lauren, and my fiancee, Emily 
Vinson. This would not have been possible without your love and support. You 
have provided me with the strength and encouragement needed to bring me to 
where I am today. 



IV 



ACKNOWLEDGEMENT 



First and foremost I would like to thank and acknowledge Jeff Smith. Without a 
timely placed email, I would have never had my interests peeked by the possibility 
of working in the field of Media Forensics. 

I would like to thank Catalin Grigoras for all of his wisdom and encouragement he 
has provided since I arrived in Denver back in August of 2011. 

I would like to thank Marcus Rogers for agreeing to be a member of my Thesis 
Defense Committee as well as provide great insight into the vast world of Computer 
Forensics. 

I would like to thank Zeno Geradts for giving me the opportunity to be an intern at 
the Netherlands Forensic Institute, and providing the freedom and resources to 
work on my thesis. 

I would like to thank Bruce Koenig and Doug Lacey for providing me with an 
advanced copy of their research in the field of DC offset research. 

I would like to thank Rachel Friedman for helping me with additional research while 
working towards completing her M.S. in Forensic Science at Marshall University. 

It is such an honor to have had the support of each and every one of you. I am proud 
to have had the chance to work with all of you, and am so grateful for all the help 
you have provided me throughout the process of receiving my M.S. 



TABLE OF CONTENTS 

List of Figures viii 

List of Tables xii 

List of Abbreviations xiii 

Chapter 

1. Introduction 1 

2. Prior Research 7 

2.1 Nine Digital Recorders 7 

2.2 Audio Compression Algorithms 9 

2.3 Acoustic Consistency 11 

3. Material and Methodology 14 

4. Results 21 

4.1 SuperNote Versus MicPro 21 

4.2 DC Offset and Standard Deviation 22 

4.3 Calculations Based on Window Sizes 24 

4.4 Histograms 27 

5. Comparisons and Conclusions 30 

5.1 DC Offset 30 

vi 



5.2 Standard Deviation 33 

5.3 Histograms 35 

5.4 Conclusions 36 

5.5 Additional Notes & Future Research 38 

Appendix 

Plots and Measurements 40 

References 103 



vn 



LIST OF FIGURES 

Figure 

1 - Signal without DC Offset - iPhone 5 - Built-in - Camera App 3 

2 -Signal with DC Offset - iPhone 5 - Built-in - SuperNote 4 

3 -Alesis PalmTrack - WAV & 35 MP3 CBRs - DC Offset Means 11 

4 - Average Standard Deviation Versus Time 13 

5 - Formula for Standard Deviation 20 

6 - DC Offset Mean and in Windows - SuperNote vs. MicPro 22 

7 - SD of DC Offset Windows and QL - SuperNote vs MicPro 22 

8 - DC Offset Mean - Built-in vs External Microphones 23 

9 - SD of Amplitude - Built-in vs External Microphones 23 

10 - SD of DC Offset Windows 25 

11 - DC Offset Mean and for Windows 25 

12 - SD at 1-Minute Intervals - iPhone 5 - EarPods 26 

13 - SD at 1-Minute Intervals - iPad 2 - Built-in 26 

14 - iPhone 5 - EarPods - Waveform 27 

15 - iPhone 5 - Built-in - Waveform 28 

16 - Histograms - iPhone 5 - EarPods 28 

17 - Histogram - iPad 1 - Built-in Mic 29 

18 - iPhone 5 - EarPods - DC Offset Plots 32 

19 - SuperNote - Test 1 - DC Offset Plots 41 

20 - SuperNote -Test 1 - Histograms 42 

viii 



21 - SuperNote - Test 2 - DC Offset Plots 43 

22 - SuperNote - Test 2 - Histograms 44 

23 - MicPro - Test 1 - DC Offset Plots 45 

24 - MicPro - Test 1 - Histograms 46 

25 - MicPro - Test 2 - DC Offset Plots 47 

26 - MicPro - Test 2 - Histograms 48 

27 - iPhone 5 - Built-in - DC Offset Plots 49 

28 - iPhone 5 - Built-in - Histograms 50 

29 - iPhone 5 - Built-in - SD Windows 51 

30 - iPhone 5 - Earphones - DC Offset Plots 52 

31 - iPhone 5 - Earphones - Histograms 53 

32 - iPhone 5 - Earphones - SD Windows 54 

33 - iPhone 5 - EarPods - DC Offset Plots 55 

34 - iPhone 5 - Earpods - Histograms 56 

35 - iPhone 5 - EarPods - SD Windows 57 

36 - iPhone 4S - Built-in - DC Offset Plots 58 

37 - iPhone 4S - Built-in - Histograms 59 

38 - iPhone 4S - Built-in - SD Windows 60 

39 - iPhone 4S - Earphones - DC Offset Plots 61 

40 - iPhone 4S - Earphones - Histograms 62 

41 - iPhone 4S - Earphones - SD Windows 63 

42 - iPhone 4S - EarPods - DC Offset Plots 64 

43 - iPhone 4S - EarPods - Histograms 65 

ix 



44 - iPhone 4S - EarPods - SD Windows 66 



45 
46 
47 
48 
49 
50 
51 
52 
53 
54 
55 
56 
57 
58 
59 
60 
61 
62 
63 
64 
65 
66 



Phone 4 - Built-in - DC Offset Plots 67 

Phone 4 - Built-in - Histograms 68 

Phone 4 - Built-in - SD Windows 69 

Phone 4 - Earphones - DC Offset Plots 70 

Phone 4 - Earphones - Histograms 71 

Phone 4 - Earphones - SD Windows 72 

Phone 4 - EarPods - DC Offset Plots 73 

Phone 4 - EarPods - Histograms 74 

Phone 4 - EarPods - SD Windows 75 

Phone 3GS - Built-in - DC Offset Plots 76 

Phone 3GS - Built-in - Histograms 77 

Phone 3GS - Built-in - SD Windows 78 

Phone 3GS - Earphones - DC Offset Plots 79 

Phone 3GS - Earphones - Histograms 80 

Phone 3GS - Earphones - SD Windows 81 

Phone 3GS - EarPods - DC Offset Plots 82 

Phone 3GS - EarPods - Histograms 83 

Phone 3GS - EarPods - SD Windows 84 

Pad 2 - Built-in - DC Plots 85 

Pad 2 - Built-in - Histograms 86 

Pad 2 - Built-in - SD Windows 87 

Pad 2 - Earphones - DC Offset Plots 88 



67 - iPad 2 - Earphones - Histograms 89 

68 - iPad 2 - Earphones - SD Windows 90 

69 - iPad 2 - EarPods - DC Offset Plots 91 

70 - iPad 2 - EarPods - Histograms 92 

71 - iPad 2 - EarPods - SD Windows 93 

72 - iPad 1 - Built-in - DC Offset Plots 94 

73 - iPad 2 - Built-in - Histograms 95 

74 - iPad 2 - Built-in - SD Windows 96 

75 - iPad 2 - Earphones - DC Offset Plots 97 

76 - iPad 2 - Earphones - Histograms 98 

77 - iPad 2 - Earphones - SD Windows 99 

78 - iPad 2 - EarPods - DC Offset Plots 100 

79 - iPad 2 - EarPods - Histograms 101 

80 - iPad 2 - EarPods - SD Windows 102 



XI 



LIST OF TABLES 

Table 

1 - Ten Microphone Setups 8 

2 - Transcoding Formats 10 

3 - iOS Software Versions 15 

4 - Recording Formats 18 

5 - DS-330 DC Offset Mean - Inconsistent Environment 31 

6 - DS-330 and SME DM-40 DC Offset Mean - Consistent Environment 31 



xn 



LIST OF ABBREVIATIONS 



ADPCM 

AES 

CLA 

DC 

DSP 

DSS 

ENF 

LTAS 

MP3 

PCM 

QL 

SD 

WMA 



Adaptive Differential Pulse-Code Modulation 

Audio Engineering Society 

Compression Level Analysis 

Direct Current 

Digital Signal Processing 

Digital Speech Standard 

Electric Network Frequency 

Long-Term Average Spectrum 

MPEG-l/MPEG-2 Audio Layer III 

Pulse-Code Modulation 

Quantization Level 

Standard Deviation 

Windows Media Audio 



xni 



1. Introduction 

In the field of audio forensics, there are numerous tests that may be performed 
during the authentication of a recording. These include examination of a digital 
recording's file structure, critical listening, waveform analysis, electric network 
frequency (ENF) comparison, and a myriad of other forms of analysis. One of the 
most recent types of examination comes in the form of measuring the affect of the 
direct current (DC) on the audio signal. This effect is known as DC bias, more often 
referred to as DC offset, and from this point on only referred to as such. As per 
Federal Standard 1037C, bias is known as:[l] 

• A systematic deviation of a value from a reference value 

• The amount by which the average of a set of values departs from a reference 
value 

In the case of DC offset, the effect of the DC current on an audio recording appears as 
a negative or positive departure of the audio waveform from equal distribution 
around the x-axis, otherwise known as the absolute zero quantization level (QL). 

When an audio recording is made, the recording device must use electrical energy. 
Either an external power supply or an internal battery provides this energy, and it 
will be present to some extent in the audio recording via the current running 
through the device. This presence may manifest itself in a number of ways, among 
which include AC hum, ENF, noise, etc. One form of energy that is always present in 
a recording is the direct current running through the recording device, and exists in 



both analog and digital recordings. The addition of the direct current to the audio 
signal is referred to as DC offset, and will affect the audio waveform by making it no 
longer centered around the absolute zero QL. This will result in a positive or 
negative shift of the waveform from the x-axis, and when the global DC offset, also 
known as DC offset mean, is calculated, the resulting value will be either positive or 
negative because of this shift from the x-axis. To calculate the DC offset, we measure 
the mean amplitude of the audio recording. An ideal audio recording would have a 
DC offset mean value of zero, but this is technically impossible due to the nature of 
complex waveforms. We must work with audio recordings in an environment such 
as MATLAB to accurately calculate the amount of DC offset. For this study, QL 
values were chosen as the unit of measurement to calculate the DC offset because 
they are the smallest form of measurement available when measuring the amplitude 
of digital audio. 

Due to the nature of digital audio, the number of available QLs per sample in a 
recording is determined by the recording format's bit depth, and the amount of 
samples per second of audio is determined by the sample rate. For example, CD 
quality audio is of the WAV PCM (pulse-code modulation) format, and has a bit 
depth of 16 and sample rate of 44.1 kHz. This means that one second of audio 
contains 44,100 samples, and each sample has 65,536 possible QL values, with 
65,536 being the equivalent of 2 A 16. Because of the compressions and rarefactions 
in acoustic energy, digital audio waveforms will analogously go between the zero- 
crossing (no acoustic energy), the peak (the compression), back through the zero- 



crossing, the trough (the rarefaction), and repeat until the end of the acoustic 
energy. Therefore, digital audio must be signed in accordance to the positive and 
negative aspects of this energy. As such, the QL values will range from -32,768 to 
32767 for a 16-bit recording, and confirms that the DC offset can be either a positive 
or negative QL value. When calculating the mean amplitude, the sum of all the QLs 
is divided by the amount of samples. It should be noted that DC offset can also be 
calculated in dB and percent, but for higher accuracy and precision it is best to use 
QLs. 

In Figure 1 and Figure 2, we see an example of an audio signal exhibiting no 
apparent DC offset followed by an audio signal exhibiting a negative DC offset. We 
say no apparent DC offset because there will always be some amount of offset for a 
complex audio waveform even when DC offset correction has been applied to the 
digital audio. From this point on, a signal that has no apparent offset will be said to 
have no offset for the sake of simplicity. The signal containing no offset fluctuates 
around the absolute zero QL, while the audio signal seen in Figure 2 lies below the 
zero QL. The signal in Figure 2 oscillates in such a way that the signal is centered at 
approximately -60 QL. 



Figure 1 - Signal without DC Offset - iPhone 5 - Built-in - Camera App 



iPhoneS-Built in-WholeRecording.wav, DC Offset Mean = -62.953936 QL, SD of Amplitude = 61.960079 QL 



o 200 la-jju.a: 

<D [ 

J -20! 



lajtoMbimMi ^^■"^"-■^"'--^■"^■^■*^*-^"^'""i-^- hjyyiii limy, -^.-^ 



< 



200 



400 



600 800 

Seconds 



1000 



1200 



1400 



Figure 2 -Signal with DC Offset - iPhone 5 - Built-in - SuperNote 

So how does this affect the forensic examination of digital audio recordings? 
Because DC offset is quantifiable on a very small and exact scale, it may be useful 
during the authentication of a recording. There will be high uniformity between any 
two products of the same type due to the assembly line nature of mass-produced 
items, which dictates that any differences should be nominal between devices of the 
same model. As such, there should be some degree of uniqueness in the DC offset 
that manifests itself in these devices' recordings due to companies using different 
manufacturing processes and parts between their various audio recording products. 
In today's digital society, we are deluged with a massive amount of products that 
incorporate digital audio recording such as digital handheld recorders, tablets, 
mobile phones and video cameras. This research focuses on a subset of these 
devices, Apple's line of mobile iDevices. iDevices can be considered any of their 
products running on the mobile operating system, iOS. These include all iPhones, 
iPads, iPod Touches, and Apple TVs. This study is not concerned with the Apple TV, 
as it is not portable, and it is not capable of recording audio. 



This particular research study aims to quantify the uniqueness of the DC offset 
present in these iDevices. iOS comes packaged with the ability to record digital 
audio via the Voice Memos and Camera apps by using the built-in microphone or an 



external microphone such as the one present on the EarPods that are included with 
the iPhone 5. Along with the ability to take photos, the Camera app is capable of 
capturing video with audio and this research tested for DC offset when using the 
Camera app's video/audio recorder. There are also many third party apps and 
microphones that are capable of recording audio, and multiple ones were tested to 
see if they affect the DC offset. Similar to any number of digital handheld recorders, 
iDevices can be a source of digital evidence. This research is meant to measure and 
examine the DC offset that is present in recordings made by various iDevices, to 
determine the usability of these measurements for the purpose of forensic 
examination, and if these measurements are at all unique when compared against 
previously tested digital audio recorders. 

Test recordings were made using various apps and three different microphones. 
The DC offset values measured for these iDevice recordings were also compared 
against devices tested in previous research. To perform these tests, the DC offset 
mean and standard deviation (SD) of the amplitude were measured per recording as 
well as the DC offset and its SD in four different sized windows. Histograms were 
made of the amplitude per recording and the four DC offset window sizes. The SD of 
the amplitude was measured in 4 window sizes. In addition, the minimum, 
maximum, mean, and SD of the DC offset were calculated for all window sizes. As 
mentioned before, QL values were chosen as the unit of measure for DC offset 
values. Likewise, all SD values are measured in QLs. 



Once measured, DC offset can be used to for exclusionary purposes. It should only 
be used as such because the value is only relatively unique, and it is possible that 
other audio recorders may exhibit similar DC offset values. Therefore, we must be 
able to distinguish between inter- and intra-variability. For this study, two sets of 
inter- and intra-variability were examined. In the first set, it was necessary to 
determine the inter-variability among all tested iDevices, and the intra-variability 
between recordings made with the same iDevice. In the second set, the inter- 
variability was examined between recordings made by the iDevices and from 
devices tested in previous research. Likewise, the intra-variability was examined 
for recordings made by devices of the same make and model. As the inter- 
variability increases, it becomes increasingly easier to distinguish between 
recordings coming from different devices. In the same manner, as intra-variability 
decreases, it becomes more difficult to distinguish between these recordings. It may 
also be possible to determine if a recording has been edited by measuring the DC 
offset at frequent intervals in a recording. 



2. Prior Research 

DC offset is a relatively unexplored form of measurement when used in forensic 
analysis, and there have only been a few studies on its effectiveness when used in 
audio forensics. The primary contributors in this line of research are Bruce Koenig, 
Doug Lacey, Catalin Grigoras, Jeff Smith, and Suzana Galic Price who have had their 
research published in the field of audio forensics and DC offset research. [2] [3] 
Furthermore, a poster presentation was given by the Author at the 46 th Audio 
Engineering Society (AES) International Conference held in Denver, Colorado in 
June of 2012. [4] To date, there have been three research studies specific to DC offset 
in digital audio recordings, two published and one a white paper awaiting 
publication; all were conducted using handheld digital audio recorders. 

2.1 Nine Digital Recorders 

The first published research study, Evaluation of the Average DC Offset Values for 
Nine Small Digital Audio Recorders, was conducted by Koenig et al. The nine devices 
tested were: 

• Olympus DS-330 

• Olympus SME DM-40 

• Olympus VN-3100PC 

• Olympus VN-8100PC 

• Olympus WS-600S 

• Olympus WS-700M 

• Philips LFH0642/27 

• SonyICD-PX312 



• SonyICD-UX512 

These recorders were given all new batteries, date and time were set, all recording 
modification features were switched off, microphone sensitivity was turned to the 
highest setting, and internal memory was chosen to store the recordings. Four input 
sources were selected for each recorder: the internal microphone, two different 
external microphones, and no input by using a dummy plug inserted into the 
microphone jack. Ten tests were conducted using these settings listed in Table 1. 

Table 1 - Ten Microphone Setups 



Test 


Microphone(s) 


Audio Input 


Analyzed Length (sec) 


1 


None 


N/A 


60 


2 


Internal 


Live Male Talker 


60 


3 


SonyME52W 


Pre-Recorded Male Talker 


60 


4 


SonyME515 


Pre-Recorded Male Talker 


60 


5 


None 


N/A 


60 


6 


Internal 


Pre-Recorded Male Talker 


60 


7 


SonyME52W 


Pre-Recorded Male Talker 


60 


8 


SonyME515 


Pre-Recorded Male Talker 


60 


9 


None 


N/A 


60 


10 


Internal 


FM News Radio 


1200 



The initial test by Koenig et al. compared the DC offset mean calculated by five 
programs, and measured in multiple formats. It was found that only MATLAB and a 
WinHex script designed to analyze audio sample amplitudes were able to provide 
accurately measured values, and that the offset should be measured in QL values for 
best accuracy. Further testing revealed that SD values for amplitude of a recording 
most likely vary dependent on the audio information being recorded. It was also 

8 



found that microphone identification would probably not be possible due to very 
small variations between the average DC offset values when comparing recordings 
made with the same microphone and with different microphones. Additionally, the 
SDs of the DC offsets between the nine recorders were inconsistent. It was 
recommended by the authors that further tests were needed for recordings made in 
different environments, with longer record times, and using more microphone and 
recorder pairings. [5] 

2.2 Audio Compression Algorithms 

The poster presentation by Fuller, How Audio Compression Algorithms Affect DC 
Offset in Audio Recordings, tested the following five handheld digital audio 
recorders: 

• Alesis PalmTrack 

• Olympus DM-520 

• Olympus WS-700M 

• Tascam DR-07 

• Zoom H2 

This research tested how transcoding from 44.1 kHz/16-bit WAV PCM to the MP3 
(MPEG-l/MPEG-2 Audio Layer III) and WMA (Windows Media Audio) formats 
might affect DC offset. Adobe Audition was used for transcoding, and MATLAB was 
used to for all calculations and measurements. Three recordings were made using 
each recorder in a relatively silent environment for a total of 15 recordings. 
Audition was then used to remove handling noise from the beginning and end of the 
recordings, and for transcoding the audio files. The files were transcoded into both 



constant bit rate (CBR) and variable bit rate (VBR) MP3s and WMAs. The encoding 
settings were as follows: 

Table 2 - Transcoding Formats 



Encoding Type 


Bit Rates 


Sample Rates 


Bit Depth 


Quality Rating 


CBRMP3 


32-320 kbps 


11,025 -44,100 Hz 


N/A 


N/A 


VBRMP3 


N/A 


N/A 


N/A 


10-100 


CBRWMA 


32-320 kbps 


44,100 Hz 


16 bit 




VBR WMA 


N/A 


44,100 Hz 


16 bit 


10-98 



Because MATLAB only works with audio in the WAV format, the recordings had to 
be transcoded back to 44.1 kHz/16-bit WAV PCM. The average DC offset for each 
recording was calculated using the three DC offset means of the three original WAV 
recordings. Next, the DC offset mean was calculated for all transcoded recordings, 
and the values were plotted for each recorder and encoding type. An example can 
be seen in Figure 3 showing the three recordings' DC offset means as solid lines, and 
their subsequent values when transcoded to all CBR MP3 settings. Similarly, the SD 
of the amplitude was calculated for all transcoded recordings, and it was found 
transcoding did not affect the waveform amplitude. The SD of the DC offset mean 
values was calculated per recorder and encoding type to determine the variance 
around the average DC offset mean. The final calculation took the maximum 
difference between the DC offset mean for each pairing of recorder and encoding 
type. Upon comparison of all measurements and calculations, the following 
conclusions were presented: [6] 

• Waveform amplitude is negligibly affected by audio compression algorithms 



10 



• DC offset is slightly affected by audio compression algorithms with the effects 
increasing as the quality decreases. 

• The amount the audio compression algorithms affected the recordings varied 
between recorders 

• The effects on the DC offset by the audio compression algorithms were 
relatively small with all but one DC offset mean having a difference of less 0.5 
QL from the original WAV recordings. 

• DC offset should be used for exclusionary purposes in forensic analysis 






-0.46 



-0.47 



-0.48 - 



-0.49 



1.5 - 



-0.51 



-0.52 



in' *i j' 



i 






/ 



ItyjpW few /V 



IW^MK, 



- 



ifi\/J,t\t % V 






I 5 10 15 20 25 30 35 

Figure 3 -Alesis PalmTrack - WAV & 35 MP3 CBRs - DC Offset Means 



2.3 Acoustic Consistency 

Koenig and Lacey's most recent study, The Average DC Offset Values for Small Digital 
Audio Recorders in an Acoustically-Consistent Environment, is a follow-up to the 
research discussed in 2.1 Nine Digital Recorders. In this research, the same nine 
devices were tested, and the following conditions were employed for all recordings: 
• 30 minute recordings 



11 



• The same acoustic environment 

• The same audio information 

• Consistent microphone positions 

This is an improvement from the prior study, as a known base for comparison is 
established for all recordings made under these particular conditions. Five audio 
formats were tested, and the average DC offset mean per recorder and their SDs 
were taken for 1-, 2-, 3-, 6-, 10-, 15-, and 30-minute segments. 

The SD values for all nine recorders as well as for six of the recorders, excluding the 
three oldest, were combined and averaged, and these values were plotted over time. 
Figure 4 shows that after approximately 10 minutes the variation in the SD begins to 
level off. The authors concluded that, among the tested recorders and settings, the 
majority of the DC offset values had a very limited range of -0.59104 to 0.01604 
except for the recorders capable of recording in the DSS (Digital Speech Standard) 
format. Additionally, differentiating between recordings made using the recorders 
and settings in this range would be extremely difficult. It was also found that SD 
decreased as recording length increased, and the SD dropped more than 75% from 
the 30-second recordings to the 30-minute recordings. Furthermore, variations in 
the SD by plus or minus 1 QL were typical for approximately 68% of the recordings, 
and this percentage jumped to 99.7% for variations of plus or minus 3 QLs. Among 
the formats tested, only DSS and ADPCM (Adaptive Differential Pulse-Code 
Modulation) had SD values that remained relatively high compared to the other 
formats. It was concluded that this is probably due to being older digital audio 
formats. 

12 



1.6 



1.4 



.2 1- 



Q 



99 

•5 
= 

■jr. 



1 .0 



()> 



0.6 






0.4 - 



0.2 



Average Standard Deviations vs. Time 




• Data Points for All 9 Recorders 
■ Data I'oleks for Last 6 Recorders 



0.0 

50 60 70 

Time in Minutes 
Figure 4 - Average Standard Deviation Versus Time 



I 



I 



30 90 100 II!) 



120 



The final conclusions of this study point to previous mentioned assertions that DC 
offset should only be used for exclusionary purposes, and that DC offset values 
should remain consistent if the recording environment and settings for a recorder 
do not change. It is also noted that duplicating the conditions of a recording when 
creating an exemplar for forensic examination may be very difficult due to factors 
such as the environment, speech amplitude, the original recorder, location of sound 
sources, etc. The authors recommend that further studies be conducted in different 
environments, enabling various recording features such as voice activation, using 
different recorder and microphones, changing the placement of the source and/or 
recorder, using multiple copies of the same make of recorder. [7] 



13 



3. Materials and Methodology 

To determine the uniqueness of DC offset among iDevices, multiple models were 

selected for this research, and multiple input sources and apps were used to make 

the audio recordings. Three mono input sources and seven apps were chosen to test 

their possible affect on the DC offset present in seven iDevices. In addition to each 

iDevice's built-in microphone, the EarPods included with the iPhone 5 and the Apple 

Earphones, both employing a remote and microphone, were chosen as external 

microphones. Additionally, a third party electret microphone, the Olympus ME15, 

was chosen for testing, but could not interface correctly with any of the iDevices, 

and was excluded from further testing. The seven selected iDevices were as follows: 

iPhone 5 

iPhone 4S 

iPhone 4 

iPhone 3GS 

iPad2 

iPadl 

iPod Touch 2 nd Generation 

Along with the iOS included Voice Memos and Camera apps, five third party apps 

were selected for testing: 

SuperNote by ClearSky Apps 
VoiceRecord by BejBej Apps 
QuickVoice by nFinity Inc 
iTalk by Griffin 
MicPro by 24/7 Apps 

These apps were chosen for a number of reasons: 



14 



• They were free 

• They were relatively popular on the App Store 

• They could export recorded audio 

The primary reason for multiple apps was to test if any differences in digital signal 
processing (DSP) might affect the DC offset. It should be noted that certain iDevice 
models are not capable of running some of these apps due to hardware and software 
limitations. The iPad 1 and iPod Touch do not contain a built-in camera, so the 
Camera app does not appear on these models. Both the iPad 1 and iPad 2 do not 
come with the Voice Memos app, as it is not included in their iOS software bundle. 
Finally, the iPod Touch is only capable of recording audio by means of an external 
microphone, as there is no built in microphone in the 2 nd generation iPod Touches. 
Table 3 lists the various iOS software versions per iDevice. 

Table 3 - iOS Software Versions 



iDevice 


iOS Version 


iPhone 5 


6.1.2 


iPhone 4S 


6.1.2 


iPhone 4 


6.0.1 


iPhone 3GS 


6.1.2 


iPad2 


6.1.2 


iPadl 


5.1.1 


iPod Touch 


4.2.1 



Before the bulk of audio was recorded, each app was tested for any possible changes 
it might make to the DC offset. These tests revealed that both the Voice Memos and 
Camera app apply DSP at some stage in the recording process that removes the DC 
offset from recordings. Further testing of the third party apps revealed that only 



15 



SuperNote and MicPro did not use DSP to remove the DC offset. Once these results 
were discovered, the next step was to determine whether or not the DC offset was 
affected by these two apps. Initial testing revealed that both apps exhibited very 
similar results when making test recordings in the same environment and for 
approximately the same length of time. It was concluded that one app, SuperNote, 
would be selected to perform the rest of the test recordings. The following sections 
will discuss these results along with all other tests. SuperNote was selected for its 
ability to easily name and export recordings as well as having smaller file sizes due 
to a lower sampling rate of 16 kHz. Additionally, the iPod Touch was found to be 
useless for any further testing, as the version of iOS running on it was not 
compatible with any of the third party apps, and the iOS apps use DSP to remove the 
DC offset. 

Prior to recording, all iDevices were fully charged to ensure proper power 
distribution, and all apps were closed save for the one performing the recording. 
Additionally, Airplane Mode was enabled to ensure no outside connectivity could be 
made during the recording process, and Auto-Brightness was turned off. An 
approximately twenty-five minute long recording was made per each iDevice and 
microphone pairing. This recording length was chosen based on the previously 
mentioned research study where it was found that fluctuation in the SD reduced as 
recording length increased, and the SD dropped significantly for recordings over 10 
minutes in length. [8] Accordingly, the DC offset mean for longer recordings will be 



16 



more consistent than for shorter recordings when acquired from the same recorder, 
settings, and environment. 

The recordings were made in a near-ideal acoustic environment to minimize the 
possibility of any transients, voices, external noises, etc. being introduced into the 
recording process. To achieve this, a small room was selected that contained many 
acoustically absorbent materials, the lights were turned off to avoid florescent hum, 
and studio-grade acoustic foam was used to surround the various iDevices. 
Furthermore, the iDevices were placed as close together as possible, and not moved 
during the entire recording process. For all recordings made using the iDevices' 
built-in microphones, recordings were made simultaneously. For recordings made 
using the external microphones, only three iDevices could be used at a time, as there 
were only one set of EarPods, and two sets of Earphones available for testing. The 
recordings made to determine if SuperNote and MicPro affect the DC offset were 
made using the iPhone 5 and its built-in microphone. These recordings were 
approximately 2 minutes in length, and recorded in the same conditions. It should 
be noted that some audio recording apps allow for a recording to be paused and 
continued ad infinitum, but this function was not tested due to the DC offset possibly 
being affected by handling noise and DSP. 

After the recording process was complete, the recordings were transferred to a 
Windows workstation running Medialnfo, WinHex, Adobe Audition 3.0.1 and 
MATLAB r2010b Version 7.11.0.584. Medialnfo and WinHex were necessary to 

17 



determine recording format information, Audition was needed to perform minor 
editing and conversion to WAV, and MATLAB was used for processing the WAV files 
and performing all scientific calculations. iTunes and iPhoto were capable of 
transferring the recordings made with Camera and Voice Memos, but were unable to 
transfer any of the third party apps' recordings. To retrieve these recordings, some 
of the apps were capable of creating a private network to share their recordings, 
including SuperNote, and others had to send their recordings by email. Once all test 
recordings were retrieved, their recording formats were examined in Medialnfo. 
WinHex and Audition were used to confirm these findings. The results can be seen 
in Table 4. 

Table 4 - Recording Formats 



App 


Format 


Sample Rate 


Bit Rate 


Channels 


Camera 


MPEG-4, AAC 


44.1 kHz 


63 kb/s 


Mono 


Voice Memos 


MPEG-4, AAC 


44.1 kHz 


63 kb/s 


Mono 


SuperNote 


AIFF,ADPCM 


16 kHz 


64 kb/s 


Mono 


iTalk 


AIFF, PCM 


44.1 kHz 


705 kb/s 


Mono 


Voice Record 


MPEG-4, AAC 


48 kHz 


108 kb/s 


Mono 


MicPro 


AIFF,ADPCM 


44.1 kHz 


352 kb/s 


Stereo 


QuickVoice 


CAF 


16 kHz 


256 kb/s 


Mono 



Adobe Audition was then used to truncate any handling noise at the beginning and 
end of the recordings. To prevent inaccuracies when calculating the DC offset, these 
edits were performed at or as close to zero-crossings as possible. To make sure the 
transcoding process did not affect the DC offset, a test recording was analyzed with 
Audition. This test recording had the DC offset measured in Audition prior to 
transcoding, and then measured again once transcoded. Measuring the DC offset in 

18 



Audition was necessary, as MATLAB does not work with any of the apps' recording 
formats. However, it should be noted that Audition calculates the offset value as a 
percentage, and was only used for determining if transcoding adversely affects DC 
offset. It was then determined that transcoding in Audition would not change the 
audio files, so all recordings were converted to WAV and brought into MATLAB for 
measurements and calculations. The edited SuperNote recordings were converted 
to 16 kHz/16-bit/mono WAV PCM uncompressed files. The MicPro recordings were 
converted to 44.1 kHz/16-bit/mono WAV PCM uncompressed files, as their native 
sample rate is 44.1 kHz. 

The two main types of calculation performed were mean and SD. The mean is 
simply calculated by summing all QL values and dividing by the number of samples. 
SD is calculated as the square root of the variance. Figure 5 shows the formula used 
to calculate the SD. The variance is found by taking the difference between each 
sample's QL and the DC offset mean, squaring each value, summing the resultant 
values, and then averaging the sum. Once calculated, the SD allows us to find the 
'normal' range of values for a particular set of values. When calculating the SD of 
amplitude for a recording made in a silent environment, values that fall outside of 
this range would be loud transient sounds such as claps, door slams, coughs, etc. 



19 



u = 



\hp*-rt 



Figure 5 - Formula for Standard Deviation 



A script was written in MATLAB that calculated the following information: 

Plot of the waveform 

The DC offset 

The SD of the amplitude 

The minimum and maximum amplitude 

Plots of the DC offset values for four window sizes 

o 5-seconds 

o 10-seconds 

o 30-seconds 

o 1-minute 
The mean for these four sets of DC offset values 
The SD for these four sets of DC offset values 
The minimum and maximum of these 4 sets of DC offset values 
Histogram plots of the amplitude and 4 sets of DC offset values 
Plots of the SD of the amplitude in four window sizes 

o 5-seconds 

o 10-seconds 

o 30-seconds 

o 1-minute 

All this information can be seen in the plots found in Appendix: Plots and 
Measurements. The following sections analyze the results of these tests, and make 
in depth comparisons between the various iDevices as well as the results found in 
previous studies of digital audio recorders. 



20 



4. Results 

4.1 SuperNote Versus MicPro 

To begin with, it was necessary to make comparisons between recordings made by 
SuperNote and MicPro to determine whether or not the apps would have any effect 
on the DC offset introduced onto a recording. Two recordings were made per 
recorder, approximately two minutes in length, and all truncations and transcoding 
were performed as previously mentioned in 3 Materials and Methodology. 
MATLAB was then used to calculate the DC offset and SDs, and the resulting values 
can be referenced below in Figure 6 and Figure 7. It should be noted that the 
recordings made with SuperNote were slightly shorter than those made by MicPro, 
and after truncating handling noise, the length of the second SuperNote recording 
dropped below two minutes. This resulted in the inability to calculate an SD value 
for the DC offset in 1-minute window because there was only one full window, and 
can be seen in Figure 7. Appendix: Plots and Measurements can be referenced for 
visual comparison between the plots for these four recordings. Based on the results 
of the calculations performed, it was found that there were no significant differences 
between the recordings to necessitate making further recordings with both devices. 
All subsequent recordings were made using SuperNote. 



21 




10-Second Window 30-Second Window i-Minute Window DC Offset Mean 



Figure 6 - DC Offset Mean and in Windows - SuperNote vs. MicPro 



a 30 




5-Second Window 10-Second Window 30-Second Window 1-Minute Window SD of Amplitude 

Figure 7 - SD of DC Offset Windows and QL - SuperNote vs MicPro 

4.2 DC Offset and Standard Deviation 



SuperNote-1 
5uperNote-2 
MicPro-1 
MicPro-2 



SuperNote- 1 
SuperMote-2 
MicPro-1 
MicPro-2 



22 




-Built-in 
- External 



iPhone5 iPhone4S iPhone4 iPhone 3G5 Pad 2 

Figure 8 - DC Offset Mean - Built-in vs External Microphones 




Built-in 
- External 



Figure 9 - SD of Amplitude - Built-in vs External Microphones 

Figure 8 and Figure 9 graph the DC offset mean and SD of the amplitude, and can be 
used for comparisons between the recordings made with the built-in microphones 
and external microphones. When reading Figure 8, make sure to note that the y-axis 



23 



is flipped for easier comprehension. We separate the values for the internal 
microphone from the external microphones because the external microphones may 
induce their own effects onto the recorded audio signal. When looking at these two 
figures, we see that the DC offset mean is higher for all recorders using the built-in 
microphone except for the iPhone 3GS, which is approximately 5 QL higher, and all 
SD values are lower for the external microphones. All SD values are relatively close 
except for the recordings made with two iPads when using their built-in 
microphones. This is most likely attributed to slightly higher overall amplitudes in 
these two recordings. 

4.3 Calculations Based on Window Sizes 

In addition to measuring the DC offset and SD over an entire recording, it is 
important to look at how these values change over time, and how the accuracy of 
these measurements change depending on the size of the data measured. For this 
study, calculations were made using 5-, 10-, 30-, and 60-second DC offset mean 
window sizes, whose plots can be seen in Appendix: Plots and Measurements. 
Additionally, the SD, mean, minimum, and maximum values were calculated per 
window size. Initial results found that as the window size increased, the SD 
significantly dropped, and can be seen in Figure 10. This drop in the SD indicates 
that increased window sizes can provide more accurate results. Furthermore, we 
can see that the SD dropped uniformly for all the iDevice and microphone pairings. 
When averaged, the SD of the DC offset in 1-minute windows came to 5.8148 QL. 



24 



so -' 




45 "' 




40 ■' 




35 ■' 




30 ■' 




25 -"■ 




20 -' 




15 ■' 

II III 




::0aHai 


-Siltt 



>.<- 



* ** 



& ..*> 



& ..*■ 



* ** 



^ ^ 



y y </ y & y # y y y y y & jp <? «• # <*- 



9 s .y ? 









Figure 10 - SD of DC Offset Windows 




# J? .J 



,jr J$ 



r .j'jf 






^5-5econd Windows 
10-Second Windows 
30-Second Windows 

■ 1-Minute Windows 



- DC Offset Mean 
■5-Second Window 

10-Second Window 
■30-Second Window 

1-Minute Window 






^ «r a? <& jr <r # <r «r jf esr <r ^ ^ <p° «t *r £ 



\^ >> *> ?> 

p _flp _«p- _op 



# 



t t t 4> 4~ t 4° 4° 4° 4° t 4° t t 4° 4> ■$• 



Figure 11 - DC Offset Mean and for Windows 

Accordingly, the minimum and maximum values for the DC offset per window size 
decreased as the window size increased. In addition, the mean of the DC offsets per 



25 



window size stayed very close to the DC offset mean value per recording except for 
the 5-second window values. Figure 11 shows of these results with the x-axis going 
from lowest to highest DC offset mean. This tells us that measuring the DC offset in 
windows will result in relatively consistent values independent of window size. 

The SD of the amplitude for these recordings stayed relatively consistent, as can be 
seen in Figure 12, where the SD fluctuates around the average SD value of 54.8000 
QL by a few QLs when measured in 1-minute windows. Analogous results were 
found for the majority of the recordings, and only the recording made using the iPad 
2 and built-in microphone had much larger fluctuations, as seen in Figure 13. This 
may be attributed to the amplitude of the audio signal slightly decreasing over time. 
Further examples for all recordings, including 5-, 10-, 30-, and 60-second window 
plots, can be observed in Appendix: Plots and Measurements. 



1 Minute Windows 

Figure 12 - SD at 1-Minute Intervals - iPhone 5 - EarPods 



O 80 — 

Q 7„ — 



I [ I i 

2 4 6 8 10 12 U 

1 Minute Windows 

Figure 13 - SD at 1-Minute Intervals - iPad 2 - Built-in 



26 



4.4 Histograms 

Histograms were made for the amplitude and for the DC offset mean per window 
size. Correlations can be made between the minimum and maximum QL values for 
these sets of data, by comparing the histograms seen in Figure 16 against the values 
found in Figure 18 for the iPhone 5. Furthermore, the minimum and maximum QL 
values of each recording are included with the histogram of the amplitude, and 
correlate to the plot of the waveform as seen in Figure 14. Like results can be seen 
when comparing the plots indexed in Appendix: Plots and Measurements. 



iPhone5-Earoods-WlioleRecording.wav, DC Offset Mean = -62.032046 QL, SD of Amplitude = 54.000031 QL 



"lilll ill 



wTTnWn., 
'JJiiiJijiji^lJjLJ 



m 

| -100 

E -200 

< 500 

Figure 14 - iPhone 5 - EarPods - Waveform 



Tin" ""r 
ll'ilLlllL 



]iyn ri ii~nnT™ 



1000 



1500 



Seconds 



When viewing the waveform seen above in Figure 14, we can observe that it does 
not oscillate uniformly. A more obvious example can be seen in Figure 15 of a 
shorter recording made by the iPhone 5 using the built-in microphone, where there 
appears to be two main QL distributions. The histogram plot of the waveform, as 
seen in Figure 16, verifies this irregular fluctuation, and shows us two primary 
peaks. Depending on the iDevice and microphone pairing, the histogram plot of the 
amplitude changes between the various recordings. The recordings made using the 
built-in microphone all exhibited a histogram with a wide distribution rather than a 
strong peak, as seen in Figure 17. Only recordings made with the external 
microphones exhibited a histogram containing two peaks, and not all iDevices 



27 



revealed these results. Both the iPhone 4S and iPad 2 had wide histograms for all 
three microphone pairings. These results can be further observed in Appendix: 
Plots and Measurements. It should be noted that test recordings were made to 
examine if these irregular fluctuations in the waveform oscillation remain in 
recordings where higher amplitude information is being recorded, but it could not 
be determined due to the density of the waveform. 



a 400 
^ 200 
"5 
S -200 
E -400 
< 



iPhone5-SuperNote-Builtln-1.wav, DC Offset Mean = -65.317363 QL, SD of Amplitude = 55.486578 QL 



1- y y 1 1 1„™„™^™„„^™„„^™„„^„ 



50 
Seconds 



60 



70 



10 20 30 40 

Figure 15 - iPhone 5 - Built-in - Waveform 

x 10 s iPhone5-Earpods-WholeRecording.wav Histogram, Min = -201.000000 QL, Max = 75.000000 QL 

S3 10 



90 



100 



S 5 



0' — 

-500 



_L_ 



_J_ 



_L_ 



_J_ 



-400 



-300 



-200 



-100 




QL 



100 



200 



300 



5-Second Windows Histogram 



400 



50C 



10 



■uV\JjJw*m-~l — > n jw^iil^A^.^ J ii*..«AJ-'iiJk.A,-.^J .A^j^,AyyV\j 



-150 



-100 



-50 

QL 

10-Second Windows Histogram 



50 



10 - 



ji ^Mi^jnuy^^^ . . 



-150 



-100 



-50 

QL 

30-Second Windows Histogram 



50 



10 



A /U.^ A ArAMJ 



-150 



-100 



-50 

QL 

1-Minute Windows Histogram 



50 



10 



-150 



-100 



-50 
QL 



Figure 16 - Histograms - iPhone 5 - EarPods 



50 



28 



s 10 



x 10 



iPadl-Built in-WholeRecording.wav Histogram, Min = -533.000000 QL, Max = 333.000000 QL 



% 5 



-500 -400 -300 -200 -100 



Figure 17 - Histogram - iPad 1 - Built-in Mic 



o 

QL 



100 200 300 400 



500 



29 



5. Comparisons and Conclusions 

For the following sections, refer to Appendix: Plots and Measurements for easier 
comparison between the recordings' plots. 

5.1 DC Offset 

The most apparent similarity between the DC offset mean values is that they all 

occur between -70.4947 and -60.0564 QL, and as mentioned before tend to be 

higher for the built-in microphones. In the studies described in 2 Prior Research, 

DC offset mean values were measured for a total of 14 recorders, with values 

ranging from approximately -64.538 QL to 23.444 QL between them.[9][10][ll] 

This establishes a baseline for comparison, and lets us know that there can be 

overlap between various types of digital audio recorders. However, while this range 

technically encompasses multiple devices, it is actually the result of multiple test 

recordings made using an Olympus DS-330 in the DSS format.[12] This particular 

recorder, as well as the DSS format, exhibits an extremely wide range of offset 

values between its various recording settings, while other devices tested in these 

different studies were much more consistent and actually comprised a much smaller 

range. Excluding the DS-330 and Olympus SME DM-40 recorders' DC offset means 

when using the DSS format, the range dropped between -10.2495 and 11.30 QLs for 

all other recorders tested in these studies. The following two tables show the 

results of this previous research found by Koenig et al. for these two recorders using 

the DSS format with the results outside the non-DSS range highlighted in 

30 



red. [13] [14] The majority of the DC offset mean values for these devices varied even 
less, and fell within a range of less than 5 QL values of QL.[15] [16] [17] 



Table 5 - DS-330 DC Offset Mean - Inconsistent Environment 



Test/Mode/Mic 


DC Offset Mean 




Tl/DSS-SP/NoMic 


-5.575 


Tl/DSS-LP/NoMic 


-1.292 


T2/DSS-SP/IntMic 


15.787 




T2/DSS-LP/IntMic 


-64.538 


T3/DSS-SP/ME52WMic 


16.062 


T3/DSS-LP/ME52WMic 


-20.591 


T4/DSS-SP/ME51SMic 


22.307 


T4/DSS-LP/ME51SMic 


-37.458 


T5/DSS-SP/NoMic 


-5.535 


T5/DSS-LP/NoMic 


-1.260 


T6/DSS-SP/IntMic 


12.914 


T6/DSS-LP/IntMic 


4.201 


T7/DSS-SP/ME52WMic 


6.932 


T7/DSS-LP/ME52WMic 


23.444 


. 


T8/DSS-SP/ME51SMic 


6.662 


T8/DSS-LP/ME51SMic 


0.054 


T9/DSS-SP/NoMic 


-5.516 


T9/DSS-LP/NoMic 


-1.256 


TlO/DSS-SP/IntMic 


-5.601 


TlO/DSS-LP/IntMic 


-10.919 



Table 6 - DS-330 and SME DM-40 DC Offset Mean - Consistent Environment 



Recorder 


Mode 


DC Offset Mean 


DS-330 


SP 


4.21283 


LP 


-34.28392 


SME DM-40 


SP 


5.78800 


LP 


-36.38407 



31 



Along with the average DC offset for the entire length of the recording, we must look 
at it over shorter periods of time, as it is possible that a recording coming in for 
forensic examination may be closer to one minute rather than twenty or more. 
Figure 18 shows how the DC offset varies between the four different window sizes. 
Upon inspection of these plots, as well as the other plots found in Appendix: Plots 
and Measurements, it becomes apparent that there is variation in the DC offset 
over time despite the relative consistency in the amplitude of the audio waveform. 

-j iPhone5-Earpods-WholeRecording.wav, DC Offset Mean = -62.032046 QL, SD of Amplitude = 54.800031 QL 



in iii j.iiii mil 



' 'IL... ILL l..JJ.L,l.UlJ..IililI.JL.,li.LJli.l.iJi 



O) 

1 -100 

E -200 

< 500 1000 1500 

Seconds 
5-Second Windows, Min = -125.771406 QL, Max = -1 .966005 QL.Mean = -60.068622 QL, SD = 39.027810 QL 
ZT -20 I — -rt- 1- "^VH - n 

o 
o 

Q -120 1-^--- ■ui-.-.r..! nj._..^.._ iju-..? 

150 
frame index 
10-Second Windows, Min = -100.809259 QL, Max = -13.949737 QLMean = -62.027832 QL, SD = 15.045847 QL 




O -50 



o 

Q . 


100 


ZT 

O 


-50 
-60 


o 

Q 


-70 


zr 
o 

o 

Q 


-55 
-60 
-65 
-70 



T" 



"T" 



T" 



"T" 




T 



bVH^ 



_i_ 



_l 



20 40 60 80 100 120 140 

frame index 
30-Second Windows, Min = -79.612999 QL, Max = -48.247150 QL.Mean = -62.006933 QL, SD = 8.112658 QL 




15 20 



25 30 

frame index 
60-Second Windows, Min = -73.027464 QL, Max = -50.080602 QL.Mean = -61.854059 QL, SD = 6.164535 QL 



45 50 




frame index 



Figure 18 - iPhone 5 - EarPods - DC Offset Plots 



When observing the amount of fluctuation between these windows in Figure 18, we 

see offset values ranging between -125.8323 QL and -2.9099 QL for the 5-second 

windows, and the range is reduced between -69.3397 QL to -46.8481 QL for 1- 

minute windows. Despite this wide fluctuation, the different windows' mean values 

32 



remains relatively consistent between the various iDevice and microphone pairings 
when compared to the corresponding DC offset mean of the recordings' amplitude. 
This can be confirmed by the graph seen in Figure 11. 

This points us towards the observation that larger windows provide stronger 
results. However, there is most likely an ideal window size. When comparing the 
DC offset mean values for all recordings against the mean for their correlating 
window sizes we find that there is very little variation among the values. The 
largest difference found was approximately 6 QL between the DC offset mean of the 
iPad 1 and Earphones pairing and the 5-second window mean value. Two other 
differences were found of approximately 5 QL, and occurred between the DC offset 
mean of the iPhone 5 and Earphones pairing, and the DC offset mean of the iPhone 
3GS and EarPods pairing when compared against their respective 5-second window 
mean values. This can be confirmed in Figure 11 as the line graph of the 5-second 
window mean values significantly deviates from the other four lines. With this in 
mind, it becomes apparent that we should use window sizes greater than five 
seconds. When looking at the difference between the DC offset mean and the other 
three window sizes, we observe a much smaller difference with the largest value 
being 0.5338 QL when comparing the DC offset mean of the iPad 1 built-in 
microphone recording against the 1-minute window mean value. 

5.2 Standard Deviation 



33 



In addition to measuring the DC offset for a recording, it is also very important to 
measure the SD of this offset as it changes over time and the SD of the amplitude. 
Calculating the SD of the DC offset over time allows us to observe the amount of 
intra-variability that occurs in the DC offset of the recording, and lets us determine 
the usefulness of the DC offset value for forensic examination. If intra-variability is 
low, then the DC offset will remain relatively consistent throughout a recording. 
This allows the DC offset mean value to be useful in forensic examination as the 
value should be consistent for all recordings made by a particular recorder and 
microphone combination. DC offset may not be useful for a recording that exhibits 
high intra-variability throughout a recording, as the DC offset amount may 
irregularly fluctuate within a wide range of values. The SD of the amplitude was 
calculated to corroborate that the recordings were all made under the same 
laboratory conditions. Therefore the range of the recordings' amplitudes should all 
be relatively close, and their SDs should be very similar in value. It is recommended 
that SD of the amplitude be excluded from forensic examination due to its 
dependence on the recorded audio signal. 

In 4.3 Calculations Based on Window Sizes, it was observed that the SDs of the 

four DC offset window sizes decreased as the window size increased, and this was 
consistent for all iDevices. It was found that the average SD of the 1-minute DC 
offset windows came to 5.8148 QL, and the range of these values was spread 
between 3.2788 and 9.5896 QL. This consistency indicates that the intra-variability 
of each iDevice's DC offset should remain relatively low, and as such makes the DC 

34 



offset mean useful in forensic examination. In correlation with the values measured 
for the DC offset mean and average per window size, as seen Figure 11, it appears 
that the 1-minute window size may be the most valuable in forensic examination, 
and that the 5-second window size is too small to provide useful results. It was also 
observed that the SD values when using the Earphones were higher for all 
recordings except for the iPad 2. This indicates that the Earphones may have the 
most adverse effect of the DC offset independent of the recording device. 

In previous studies, the largest SD of the DC offset mean values was found to be 
19.49 QL for the Olympus DS-330.[18] This indicates that the DC offset in iDevices is 
significantly different than that found in these previously tested digital audio 
recorders. Furthermore, all other SDs found in these previous studies were much 
lower than those found for the Olympus DS-330. Unfortunately, these values were 
calculated in a different manner than in this study, and as such cannot be directly 
compared to the results found here. 

5.3 Histograms 

In Appendix: Plots and Measurements, we can observe and compare histogram 

plots from all the recordings. These histograms provide visual correlation between 

the DC offset, and the SD. All amplitude and DC offset window histograms have the 

same respective X and Y scale so they may be viewed and compared with greater 

ease and in equal proportion. The histograms that contain two peaks correlate with 

the waveforms that appear to have two main QL distributions such as that seen in 

35 



Figure 16. As mentioned before, the histograms that do not have two peaks have a 
wide distribution rather than a strong peak. 

The most apparent observation that can be drawn from the DC offset window 
histograms is the relatively wide dispersion of the values. For all window sizes, 
there are no particularly strong reoccurring values. However, as the window size 
increases these values become less spread out, and tighten around the DC offset 
mean value. Correlation can be seen between this trend and the corresponding 
plots of the DC offset windows and their minimum, maximum, mean, and SD values. 
Furthermore, the amplitude histograms, which contain two peaks, are relatively 
equally spread around the DC offset mean value. These histograms containing two 
peaks correlate with the waveforms that have two strong QL distributions, and 
examples can be seen when comparing the plots for the iPhone 5, 4, and 3GS and 
iPad 1 when using the external microphones. 



5.4 Conclusions 

The measurements and comparisons conducted in this research point to a few main 

conclusions regarding the use of DC offset in forensic examination, and whether or 

not iDevices exhibit any unique traits when compared with other digital audio 

recorders. With respect to previous findings that DC offset traits can be similar 

across multiple recording devices, it is still recommended that any measurements 

be only used for exclusionary purposes. Furthermore, when used in the forensic 

36 



examination of audio, these tests should only comprise a part of the analysis, and 
many other forms of inquiry should be performed such as spectral analysis, 
waveform analysis, ENF analysis, long-term average spectrum (LTAS), compression 
level analysis (CLA), etc.[19] 

For all tested iDevices, it was found that the DC offset mean remained relatively 
consistent between the various iDevice and microphone pairings. Similarly, the SD 
of the amplitude for these recordings fell within a comparably small range, save for 
the iPad 2 when using the built-in microphone, which indicates that all recordings 
had very similar recorded audio signal amplitudes. When measured in 1-minute 
windows, it was found that the SD of the DC offset had very minor variations among 
the tested iDevices. These findings lead to the conclusion that there should be 
relatively low intra-variability of the DC offset values between recordings made by 
the tested iDevices. When compared with measurements taken from previous 
research, we can conclude that these iDevices are relatively unique as there is nearly 
zero overlap when comparing DC offset and SD values. It can also be said that while 
having a relatively low intra-variability between iDevices, there is a high inter- 
variability when compared to other devices. Furthermore, the low intra-variability 
of the DC offset mean and SD values will likely increase as the recording length 
becomes shorter. Finally, it should be noted that certain iDevice and microphone 
pairings might be more identifiable when analyzing the histogram of their QLs. 



37 



5.5 Additional Notes & Future Research 

While many of the apps tested in this research used DSP to perform DC offset 
removal, they still exhibit an extremely small amount of offset due to the complexity 
of the resulting recorded waveform. There may also be other visible effects of the 
recording process when viewing the waveform. For example, a video recording 
made with the Camera app will result with an audio waveform that does not actually 
begin at the first sample. In effect this means that the waveform remains at a 
constant QL until the app starts feeding audio information into the video 
recording. Such manifestations may be useful when trying to identify a particular 
recording, and further research needs to be conducted on this issue. 

As with any research, there is always the need to conduct more studies. This is 
especially true for DC offset, as it is a relatively unexplored form of measurement 
when used in the forensic examination of digital audio recordings. One of the most 
important parts of the research of DC offset is making test recordings with as many 
devices as possible, and there are a plethora of devices that have yet to be tested. 
This is compounded by the possibility of numerous recording settings and formats 
per recording device. Since a large portion of the population owns mobile phones 
and other devices capable of recording audio, it is important to test devices that are 
not typical handheld digital audio recorders. In addition to the iDevices tested in 
this research, there are still many more that are capable of recording audio as well 
as a numerous other smart phones. 



38 



Along with the need to test new devices, real world examples need to be taken into 
account. Ideal conditions will not be the norm when digital audio evidence is being 
tested in a forensic lab, and research must be conducted that addresses this issue. 
One such study has been performed, but concluded that a wide variety of tests still 
need to be performed in a variety of environments, with longer recording lengths, 
and with a larger variety of recorders and microphones. [20] This thesis expands on 
these ideas by testing a number of new recording devices, and making recordings at 
longer lengths. However, research still needs to be done with recordings that mimic 
real world conditions that include room noise, handling noise, start and stops, etc. It 
may also be beneficial to test how the energy going into a recorder may affect the DC 
offset. The type of battery, the charge of a battery, and if the recorder is powered by 
an AC power supply all might impose their own effects on a recorder. 

In addition to testing new devices and in different environments, future research 
should incorporate a wider range of calculations such as those used in this study. 
The use of histograms can help analyze the range of DC offset values, and better 
visualize the waveform for comparison. Measuring the DC offset and SD in windows 
can let an examiner see how much fluctuation occurs throughout a recordings, and 
can be useful in comparisons between other recorders and recordings. Along with 
incorporating more calculations, the results of DC offset research should be 
aggregated for easier use in forensic audio examination. 



39 



Appendix: Plots and Measurements 

This section provides plots of the waveform, DC offset in windows, histograms of the 
amplitude and DC offset in windows, and SD of the amplitude in windows for all 
recordings. Along with these plots, various calculated values are included per 
recording. Among these are, the values for the DC offset mean, SD of the amplitude, 
the mean of the DC offset values for the various window sizes, the SD of these offset 
values, and minimum and maximum values for the amplitude and the DC offset 
values in windows. 



40 



o 

CO 

I-- 

UD 
CO 
CO 

"5f 

ud 
un 

ll 



E 
< 



Q 

to 



o 

CO 
CO 

I-- 



O 
O 
Q 



to J-j 



I ill 



n 


m 


in 






CD 




.-: 




CO 




r~- 








o 




UD 




■* 


n 


II 


CO 


Q 



'J' 



o 

UTI 

» En 
^ CO 

CD S ^ 
LD tu CM 

m ,; 

TO 



O 

CO 

CD 
O 

o 

CO 



MO 



DO CO CO CO 

oo o o 

"Td CM T 



ho] apnmdujv 



n i i i ii <-m 



J.LJ.LJ 



-i ^ 



fTT 



o 

LO 



II 

Q 



o 



IT) 



o _; 



CO 

m 
o 



Cf 

CD 
CD 

ijn 

Lfi 

CD 







co 
o 



CM^OLOIOCDrN 



ho] oa 



n- 



J I L 



a 



CD 
II 

Q 
CO 

_F 

a 



in 
co 



C3 





m 


'1' 


1. 1 
CO 


c 


m 


'1' 


^~ 


■t- 


LU 








II 



CO 

■.-■ 

CO 



ho] oa 



t — i — r 



j i L 



o 



o 

CD 
CD 
CD 
II 

Q 
CO 

_F 

'2' 



LO 

'3". 



CN 



Cf 
CO 

-5 LO 

a) CD 
E LO 

£ II 



CO 
CO 

.:■ 

CD 
CO 
CD 



^ II 



ho] oa 



i i i i H 



_ j _ i _ 

- n - T - 
_ j _ i _ 



-i ^ 



j -=r 



I 



ho] oa 



Figure 19 - SuperNote - Test 1 - DC Offset Plots 



41 



a 

o 
o 
o 



o 

o 
o 



DD 









o 



o 






- o 



a 



01 



_L 



o —i 

Lf) O 






CN ^- OLT1 

sa3uajnooo# 



o 

ld 0^7 
saouajnooQ# 



\ 




P r 



O —I 
Lf) o 



o 

saouajnooo# 




II L 



o —i 

Lf) O 



o 

cm ^- 0^7 

saouajnooo# 






o —i 

Lf) o 



Lf) 

o 



saouajnooo# 



Figure 20 - SuperNote - Test 1 - Histograms 



42 



o 

(7) 



o 
o 



E 
< 



Q 

to 



o 



o 

CD 

-st- 
un 
CM 



o 
o 

Q 






to 



F 



o 



o 



n 




_l 


LU 




(J 

o 
n 












m 






i-r. 
















m 


-3- 




LVJ 


II 

X 


n 




^ 


UJ 




_r 
o 

CO 

CO 



oo o o o 

"TM CM T 



ho] apnmdwv 



I I I IT 



£ : :L, 



I I III 1 1 



o 



■3 



o 



ho] oa 



TTI 



L_L 



a 
o 

CD 



II 
Q 

co 

_r 

a 

CD 

r-- 

LD 
O 
00 



Cf 





m 


.K. 


in 


<D 




<J 


t 


C 


|. 1 




HI 






E 


"3- 



a 



CD 
i-.-i 

en 



CO 

o 



ho] oa 



i — r 



<3 
en 



en 
cr> 
II 

Q 
CO 

_r 

a 



CD 



O 

* cd- 
unl ffl 

fN 0i CO 

E ^ 



a 



[10] oa 



;_ cj) 



j_ ■* 



U L 



ho] oa 



Figure 21 - SuperNote - Test 2 - DC Offset Plots 



43 



cs 

o 

o 

o 
o 



O 
o 
o 

o 



op 






o 
a 






_L 



<TN ^- OLD 

sa3U8jnooo# 



i r 



V 



< 



I 



J L 



O —I 



Op 
o 



\ 



o —I 



Op 

o 

,-■-, 



"i r 




j L 



o —I 






o —I 



o ld 0^7 

saouajn330# 



ld 0^7 

S83uajnooo# 



S8ouajnooo# 



S8ouajnooo# 



Figure 22 - SuperNote - Test 2 - Histograms 



44 



o 



co 
cm 

to 



E 
< h 



Q 

to 



o 

CM 
CO 
■" 1 
CO 
CT) 

cd 

ID 



o 
o 

Q 



o 





CO 
CO 




CD 




*=r 


CM 


cm 

CO 


^~ 


II 




n 




UJ 




_i 
o 




o 

CM 


o 


CM 



o 



03 
CO 

ID 



m co 
™ li 



O 



°? 



oo o o o 

DO CO CO 

"TM (NT 



ho] spnmdwv 



Til I I 



J 



o 

IT) 
LO 

" I 
CO 
i-.-i 



II 
Q 

co 

_r 

a 

CM 
CO 

'" I 
CO 
CD 

CO 



Cf 

CO 

CD (XI 
"a CM 



^ E 



a 



o 
o 

CO 



co 
o 



i i i r 










r-rMrr 


! ! ! 


o o o o 

CM T CO CO 



III I I I 



a 

CO 

CD 
LO 

-=r 

CD 



II 
Q 

co 

_r 

a 

CM 
CO 

'" I 
CO 
L7> 

CO 
CO 



(3 





■=t 


.K. 


i i 


<D 


■=r 


<J 


i M 


■T" 






I M 






E 


LO 



a 



CO 
LO 



CO 

o 



r-r *•--,- CM 



ho] oa 



ho] oa 



i 1 1 i 



UO 


a 


^J- 


CO 




'-■■ 




T 




CD 




,-.-, 




CO 




CO 




II 




n 




CO 


T 


_i 




tj 




t 




*=r 




^t 




CD 




'- 




CD 




ao 




LO 


in 


ll 


CO 


c 



— CO 



a 

^ LO 

- oq 

CD CO 

F lo 



O 



CO 

II 



ho] oa 



__L CD 



_J_ T 



ho] oa 



Figure 23 - MicPro - Test 1 - DC Offset Plots 



45 



o 



o 
o 
o 



O 

o 
a 
o 
a 
o 



o 
a 




o 
-- o 



_L 



<TN ^- OLD 

sa3U8jnooo# 



i r 



J L 



O —I 



m 
o 



\ 



O —I 



o 

,-■-, 




o —I 



I 



o —I 



saouajn330# 



LTl 0^7 

S83uajnooo# 



01 o^ 
S8ouajnooo# 



S8ouajnooo# 



Figure 24 - MicPro - Test 1 - Histograms 



46 



o 

CD 

CO 
CO 



E 
< 

•■*— 

o _ 
Q 

m 



o 



o 

CO 
CM 



II _ 



■<t= 

o 
o 

Q 






~T~1 



o 



-st- 
un 



II 

Q 



o 



o 

CD 



-3 & 

^ CO 

n to 



10 CO 



o 









DD CD CD CD 

"TM (NT 



ho] apnmdwv 



iU — . 



I I I I I II CM 



I 







o 



II 

Q 



O 4 






X CO 
03 U> 



a 



CD 

'=■ 
P-! 



S J 






l>WLODCOrM 



ho] oa 



. . _ _ J 1 — 



a 



II 

Q 
CO 

_r 

a 



to 
-3- 



<3 



E ^ 



a 



ho] oa 



J L 



a 

un 
o 

h~ 
o 

CN 
II 
Q 



a 



LO 
-rr 



O 
o 

N aj od 
E ^ 

£ ii 



a 



ho] oa 



1 L 



ho] oa 



Figure 25 - MicPro - Test 2 - DC Offset Plots 



47 



cs 

o 

o 
o 
o 



cs 

o 



] 



a 



m 



<TN ^- OLD 

sa3U8jnooo# 



( 

T 
! 

( 



O —I 



US 



i r 




y 



J L 



O —I 



o 

CO 



"i r 



J L 



O —I 



LTl 0^7 



S8ouajnooo# 



01 ■>- 0^7 
S8ouajnooo# 



S8ouajnooo# 



Figure 26 - MicPro - Test 2 - Histograms 



48 



o 

CD 

t-~ 

CO 
(J) 

s 

II 



E 
< 



Q 



o 

CD 

ITS 
'-■ 

n 
trs 



O 

o 

Q 



CD 
Lf) 



O O O 
CM CN 



O 
CD 
CO 



II 
Q 

m 

_r 
o 

01 
CO 

o 



O 
n 

-g ^ 

^ CM 

° CM 
U ■ 

a: CD 
™ li 

05 



a 



CO 




j: <m e 



ho] spnmdwv 



ho] oa 



ho] oa 



ho] oa 



ho] oa 



Figure 27 - iPhone 5 - Built-in - DC Offset Plots 



49 



o 

o 
o 
id 

CM 



a 

o 
o 

CTJ 



£ — 



00 

LT1 






O 



o 

o 

old 



1 T 



J L 



o —I 



o 



TT 



J L 



o —I 



o 

,-■-, 



"i r 



j L 



o —I 



i r 



j L 



o —I 



sa3uajno30# 



saouajn330# 



saouajn330# 



sa3uajn33Q# sa3uajn33Q# 



Figure 28 - iPhone 5 - Built-in - Histograms 



50 











3 

o 
o 

ct: 
o 


— I^S" 


'5 

CO 

a> 

c 
o 

CL 


^™^ i 



O 



O 
<_> 

00 



Tlrg 



o 



o 
<_> 

Q> 

00 



o 



o 



-.- Ul 



O 
(_) 
Q) 

00 
o 



-J- LD 



ho] as 



ho] as 



ho] as 



[id] as 



Figure 29 - iPhone 5 - Built-in - SD Windows 



51 



■■ 

CO 
PI 

n 



E 
< 



Q 

to 



<3 

CO 
CD 

:c 
n 

ID 



CO 
II 



O 
O 
Q 



CC 



LU 
Lf) 



J I I 



ZZ=? 



o 





CD 




, - 




r- 




CM 




CD 


a 


T 


cj 
cm 










II 




n 




to 




_i 




<j 




r~- 




-3- 






i i 




i i 




LJ 


CM 








CO 




LD 





o ° 

CO CO 

^-g cS 

5 5 



a: 

w 



C' 



CO 

'.O 

co 

CO 
CM 





ss 




O 



CO 



ll 

Q 
CO 

_r 

<:■' 



cm 



a 

o 1? t^ 

to ■- ro 

E ™ 

£ N 
>=: 
en 



OS 



a 

cm 

LO 
r.-i 
■=■ 
h~ 

*=r 

CM 

II 

Q 

m 



O' 



CQ 
II 



a 

CO x l^ 

C- 1 ai CO 

"a CO 

— CD 

TO ' 

■^= II 



O 





CD 




'JJ 








CD 




Ln 




■=t 




CD 




CO 




II 






o 




■* 


.£ 



rn r 



Jj_l L 



CO 



CO 

II 
a 

CO 



o - 



CO 



CO 

ll 



D 

o 



_;--,_i-;_ co 



a - 



.1. '~. 

is 

LD- N 

E T 
£ N 



a 



GO 



ft 



L--1- CM E 



ho] apnmdujv 



[ID] DO 



[ID] 3d 



[ID] oa 



[ID] 3Q 



Figure 30 - iPhone 5 - Earphones - DC Offset Plots 



52 



a 

o 
o 

O 



o 
o 

rn 

CD 
■" I 



LU 



1 



a 



"i r 



j L 



O —I 



o 



"i r 



J L 



"i r 



J L 



O —I 



i r 



j L 



o —I 



o un old 

sa3uajno30# 



o Ln 0^7 
saouajn330# 



saouajn330# 



sa3uajn33Q# sa3uajn33Q# 



Figure 31 - iPhone 5 - Earphones - Histograms 



53 



o 
o 

o 



(u 

c 
o 

JZ 
Q. 

LU 

LD 
(U 

C 

o 



j-^^^^P- 



o 



o 

CM 



o 



O 
<_> 

Q> 
CO 



LO 



o 

00 j= 



o 
<_> 
CD 

00 



\n i r 






o 



o 

!_) 

CO 
o 



o 



ho] as 



ho] as 



[id] as 



U I U 

o o o 

(JD -st CM 

[id] as 



Figure 32 - iPhone 5 - Earphones - SD Windows 



54 



f--^ 



o 

o 

.■■. 



E 
< 






o 

cm 
m 
o 



O 
O 
Q 



LU 



O 



o 



a 



C-' 



CO 

o 



ll 
a 

CO 

_f 

a 

rn 

CO 



O 



1§ 

.- CD 

a. ro 
E v 

>=: 



Cf 



■3". 
CD 



ho] apnmdujv 



rjfltfcaDrN 



ho] oa 




ho] oa 



[id] oa 



[id] oa 



Figure 33 - iPhone 5 - EarPods - DC Offset Plots 



55 



o 

o 
o 



o 



LU 



O 

a 




o 
-- o 



_L 



_L 



o —I 



o 



_L 



o —I 



o 

,-■-, 



_L 



o —I 



1 T 



J L 



o —I 



o un old 

sa3uajno30# 



saouajn330# 



saouajn330# 



O 0^7 O LJTi 0^7 

sa3uajn33Q# sa3uajn33Q# 



Figure 34 - iPhone 5 - Earpods - Histograms 



56 



c 
TJ 

o 
o 
(U 

ct: 
(u 
o 

? 

J? 

"O 
O 
Q. 

nJ 
LU 

iA 
<D 

c 
o 

.c 

CL 


^v 




^E 1 



% 

o 



O 
<_> 

00 



o 

"CS 



o 
<_> 
CD 

00 



TlrM 






<» o 

(_) 

Q) 

00 
o 



-1- ^n 



o 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 35 - iPhone 5 - EarPods - SD Windows 



57 



o 

CD 
CM 

r~- 

CO 



E 
< 



Q 
US 



o 

CD 



LD 



o 
o 

Q 






o 

CO 

,-■-, 

ijO 
CD 
LD 

en 
II 

Q 

m 



o 
<tm 

CO 

CM 

o 

LD 
LD 



o 



LD 



O 
LD 



O 
LD 



W 
LD 



DDDOD 



ho] apnmdujv 




FT 



J 



ho] oa 



ho] oa 



ho] oa 



i ii i 



ho] oa 



Figure 36 - iPhone 4S - Built-in - DC Offset Plots 



58 



o 

o 

o 
o 



O 

o 
o 

r-- 

r-j 









o 



o 

o 

old 



_L 



o —I 



o 



_L 



o —I 



o 

,-■-, 



"i r 



j L 



o —I 



1 T 



J L 



o —I 



sa3uajno30# 



saouajn330# 



saouajn330# 



i-n o Ln o^- o uri o^- 

sa3uajn33Q# sa3uajn33Q# 



Figure 37 - iPhone 4S - Built-in - Histograms 



59 



c 

T3 
O 

o 

ct: 
o 

? 

'5 
GO 

CO 

c 
o 

a. 


i art i 



O 



O 
<_> 

00 



I 111 I I 




TT 



i i n ™ 



t 






o 



O 



o 
<_> 
CD 

00 



o 



o 
(_) 
a> 
00 
o 



o 



I I 



I I 



I I 



ll I I 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 38 - iPhone 4S - Built-in - SD Windows 



60 



o 

in 

CO 
CO 



ll 



E 
< 



Q 

CO 



'-> 






O 

o 

Q 



cr 



CO 



o 





en 




:■: 




CO 




*=r 




CO 


i M 


r- 


*— 


co 




II 




n 




co 




_i 




cj 




CM 




"5f 


o 


1"- 

co 


a 


-=)- 




C'S 



n 






i_j 






LU 
















rN 






r-- 
















m 






LVJ 


II 

X 


n 




_ 


i i 




_i 


LU 




un 

CM 
CO 
CM 

-5f 

CD 



CO 



oooo 

CM CMT 



ho] apnmdwv 







O' 



CD 


LO 


Lf) 


h~ 


CM 


*=T 




LO 




CM 




* — 




II 




n 




CO 




_i 




tj 




IT) 




(JJ 




Lfi 




OD 


CD 


■-" 


O 


CD 








i-n 




lO 



o 



LO 

o 



Cf 

LO 
O 
LO 



r*ttfcccrj 



ho] oa 




o 



<3 



a 



o 

CO 


"a 


CD 




m 


r— 




!- 


UJ 










*+= 


II 

X 

as 


n 




_l 


UJ 




(J 

LD 
GO 

CD 
CD 
03 



I- ° 



[ID] 0Q 



a 



-l- S 



<3 



ill 



il- !£ 



[id] oa 



--t. LO 



a 



■ ■I" E 



a 



7 

□J 



J I u 



[id] oa 



Figure 39 - iPhone 4S - Earphones - DC Offset Plots 



61 



o 

o 
o 

CM 



O 

o 
o 









o 



o 

o 

old 



"i r 



j L 



o —I 



o 



"i r 



J L 



o —I 



o 

,-■-, 



"i r 



j L 



o —I 



i r 



j L 



o —I 



sa3uajno30# 



saouajn330# 



saouajn330# 



sa3uajn33Q# sa3uajn33Q# 



Figure 40 - iPhone 4S - Earphones - Histograms 



62 













— - 


A; 








d) 

c 

TJ 

i_ 

O 

o 

<D 
Ql 
<D 

O 

.n 

J? 

c 
o 

.n 


— - 


---^H ^H 




^^V ; 


LU 
CO 

c 
o 
.n 
0. 


^^3 




^p 1 




^^^^^^^^v 




1 ^p 








T^^: 



o 



O 
<_> 

CO 



IT I 



o 



O 
<_> 

'■■'< 



i ri 



CO 

o 



O 
O) 

CO 
■ 

CD 



-T- ° 



-i- !£ 



TT r 






I I I L 



o 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 41 - iPhone 4S - Earphones - SD Windows 



63 



o 



"5f 

n 
ll 



E 
< 



Q 



o 
o 

en 
1.3 



O 

O 
Q 



ce 



GO 



o o o o 
o o o 

CM D t 



ho] apnmdwv 




f 



ho] oa 



ho] oa 



ho] oa 



ho] oa 



Figure 42 - iPhone 4S - EarPods - DC Offset Plots 



64 



cs 

o 

o 
o 

r-- 

01 
OJ 



a 

o 
o 
o 



cr 









a 



o 

a 

old 



o —I 



< 
3 



_L 



m 

o 



_L 



o —I 



o 

,-■-, 



"i r 



j L 



o —I 



1 T 



J L 



o —I 



sa3uajno30# 



saouajn330# 



saouajn330# 



i-n o Ln o^- o uri o^- 

sa3uajn33Q# sa3uajn33Q# 



Figure 43 - iPhone 4S - EarPods - Histograms 



65 



c 

T3 
O 

o 

(U 

o; 
o 

? 
w 

"O 
O 
Q. 

nJ 
LU 

CO 
^t 
(U 

c 
o 
.n 
a. 


H--Ffc-!"h 
i iB i 



<S1 

% 

o 



i- Ln "o 



o 
00 



XT 



J_L 



_L_L 



■I I I 



I III I lift 






o 



o 



o 
<_> 
CD 

00 



o 



o 
(_) 

oo 

o 



o 



.;_ o 



I- !=■ 



I I 



I I 



I I 1 II 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 44 - iPhone 4S - EarPods - SD Windows 



66 



o 



E 
< 



Q 

to 



o 

CD 

t-- 



o 
o 

Q 



CD 



OOOOO 

ho] spnmdwv 




1TT1 cm 



o 

m 

r-j 
Ln 

r-- 
m 

ll 
Q 

CO 



o 









O 



E ^ 



Cf 



T 



+ 2 



op 
II 



CO 



J U 



ho] oa 



ho] oa 



ho] oa 



ho] oa 



Figure 45 - iPhone 4 - Built-in - DC Offset Plots 



67 



o 

o 
o 



O 
o 

o 

CM 



£ — 



00 
-st 






O 



o 

a 

old 



_L 



o —I 



o 



_L 



o —I 



o 

,-■-, 



"i r 



j L 



o —I 



1 T 



J L 



o —I 



sa3uajno30# 



saouajn330# 



saouajn330# 



i-n o Ln o^- o uri o^- 

sa3uajn33Q# sa3uajn33Q# 



Figure 46 - iPhone 4 - Built-in - Histograms 



68 





- .I.^^J— . 




i^fe j 


3 


^^S| 


o 
o 

ct: 






o 


__^^Vj_ 


c 
■i-* 


_J^£^_H 


3 

CO 

c 
o 

CL 




^^--j- 




""^fci" 




l^j — i 



<S1 

% 

o 



O 
<_> 

00 




II I 



TlLD 



n n ™ 






o 



o 



o 
<_> 
<Ii 
00 



o 



o 
(_) 

00 
o 



-\---\- S 



1- CD 



I ■ I 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 47 - iPhone 4 - Built-in - SD Windows 



69 



o 

CO 



E 
< 






o 

CD 

CO 



O 
o 

Q 



LU 
■si- 



O O O 
CM CM 



ho] apnmdujv 




[ID] 3d 



[10] 0Q 



ho] oa 



ho] oa 



Figure 48 - iPhone 4 - Earphones - DC Offset Plots 



70 



o 

o 
o 

PI 

" I 

CM 



a 

o 

o 
o 



cr 



LU 






o 



o 
a 

OLD 



"i r 



j L 



O —I 



o 



"i r 



J L 



o —I 



o 

,-■-, 



"i r 



j L 



o —I 



i r 



j L 



o —I 



sa3uajno30# 



saouajn330# 



saouajn330# 



sa3uajn33Q# sa3uajn33Q# 



Figure 49 - iPhone 4 - Earphones - Histograms 



71 



3 

di 

o 
<J 
a> 

a> 

o 

.c 

w 
(u 

c 
o 
x: 
a. 

LU 

4 

c 
o 

.n 


i_ ^^^^L^. 

i V! 



O 



O 
<_> 

to 



O 



o 
<_) 
CD 

CO 



o 



o 
(_) 
a> 
CO 
o 



o 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 50 - iPhone 4 - Earphones - SD Windows 



72 



o 

n 

o 
itm 
o 



E 
< 






o 
o 

(T) 

1J3 
CO 
CM 



O 
O 
Q 



LU 



O 

r\i 
o 

CO 



II 
Q 

m 

_r 
o 

un 
li- 
en 



o 






l_J 
a 




a 






rn 












i-, 






^J- 
















m 






LVJ 


II 


i i 






n 






m 







a 

CN 

CO 

li- 



ra 



DDDOD 



ho] apnmdujv 




rj^LaroN 



ho] oa 




ho] oa 



[id] oa 



[id] oa 



Figure 51 - iPhone 4 - EarPods - DC Offset Plots 



73 



o 



o 
o 



a 

o 
o 



cr 



LLJ 






a 






un 



_L 



o —i 

Lf) O 






OLT1 

sa3uajno30# 



o 

saouajn330# 



_L 



o —i 

Lf) O 



O 

saouajn330# 



_L 



o —i 

Lf) O 



"i r 



j L 



O —I 
Lf) O 



o 0^7 O Lf) O^-p 

sa3uajn33Q# sa3uajn33Q# 



Figure 52 - iPhone 4 - EarPods - Histograms 



74 





4^p i 




















'^C»_ 






c 

o 
o 




— 


Tg; 


jd 


Tir 


u 


.^HlH 


J? 


:JB» 


O 




Q. 


-1--^^^W — 


10 


B 




LU 




c 
o 

a. 




^B^~ 










' ^Ei_L_ 




^Vj ^BH 




te^BJ^ 




b1 




MB 




i^S 





01 

R £ 
S o 



o 
00 



° -8 



o 
<_> 
CD 



_ J 



O 



o 
(_) 

CD 



o 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 53 - iPhone 4 - EarPods - SD Windows 



75 



o 



CD 
r-i 



E 
< 



Q 



o 
o 

(T-, 

o 
o 



O 
O 

Q 



m 

CO 




II I I I 



-1- CM E 



_L_Ll 



J_L 



ho] apnmdwv 



ho] oa 



ho] oa 



ho] oa 



ho] oa 



Figure 54 - iPhone 3GS - Built-in - DC Offset Plots 



76 



cs 
o 

O 



o 

o 

o 
a 
o 

oo 

■" I 



CD 

01 






a 






un 



"1 T 



J L 



o —i 

Lf) O 






2 s 



sa3uajno30# 



o 
o in 0^7 

saouajn330# 



"i r 



J L 



o —i 

Lf) O 



o 

saouajn330# 



"i r 



J L 



o —i 

Lf) O 



i r 



j L 



.i O -1 
Lf) O 



o un 0^7 o ld 0^7 

sa3uajn33Q# sa3uajn33Q# 



Figure 55 - iPhone 3GS - Built-in - Histograms 



77 







^g^ 










^^~ 










"^* 




Si 







r 


JV^^^L _ i _ 






u 

o 
o 








"5 


^^L-. 


— 


'5 
GO 

CO 

o 

c 
o 
x: 
0. 




.^^^■P 


_ 














i 


-^^| ^B - 







^fc; 






^r 



o 



O 

<_> 

(Is 

CO 



£ 
o 



O 
<_) 

CD 

CO 



LO 

o 



O 
(_> 
CD 

CO 

I 

o 



tt r~i™ 



I I I L 



o 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 56 - iPhone 3GS - Built-in - SD Windows 



78 



o 



en 
ll 



E 
< 



Q 

01 



o 

CD 

m 
n 

CD 



1= 

O 

o 

Q 



LU 

th 







1 1 1 




--L_-j---f 










--;■--;---} 




-r-Mi--f 


! ! ! 


o o o 

(I) K CO 



CM E 



_-,.--' > CD 



ho] apnmdwv 



ho] oa 



ho] oa 



[id] oa 



[id] oa 



Figure 57 - iPhone 3GS - Earphones - DC Offset Plots 



79 



o 



o 
o 
o 



a 

o 
a 



m 

CD 






O 



o 

a 

old 



"i r 



: 



J L 



. J o — i 



o 



"i r 



J L 



o —I 



o 

,-■-, 



"i r 



j L 



o —I 



i r 



j L 



o —I 



sa3uajno30# 



saouajn330# 



saouajn330# 



sa3uajn33Q# sa3uajn33Q# 



Figure 58 - iPhone 3GS - Earphones - Histograms 



80 



o 
<J 
a> 

Ql 

a> 

o 

x: 

w 
(u 

c 
o 

x: 

a. 

i_ 

(0 
LU 

CO 

o 

CD 
(U 

c 
o 
x: 
0. 


^m\ 



% 

o 



O 
CO 



E 



o 

00 *= 



o 
<_) 
05 

CO 



o 



o 
c_> 
a> 
CO 
o 



o 



II I L 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 59 - iPhone 3GS - Earphones - SD Windows 



81 



o 

CO 

t-- 



E 
< 



Q 

to 



o 

LO 

itm 

un 
o 
itm 



O 
o 
Q 



o; 



lu 
to 

IT) 



o 



CO 



II 

Q 

to 

_r 
o 
un 

*=r 

LQ 



O 

CD t— 

m li 

03 



O 



o 


_l 

O 




o 
,-.-, 

■* 




LO 




h- 




II 


o 


Q 
CO 




_f 

O 




CD 

r-i 

'- 




* — 


O 


01 
LO 

II 







o 



01 

CO 
LO 

II 

Q 
CO 

_f 

o 

LT> 



o 




a 






H 




"a 


CN 




ai 


■=t 


r-j 


TO 


L1J 
II 



(3 



o 

CO 



t— "a 



CO 



'I I I I 



^--i-HR 



t 



I I I II 



ho] apnmdujv 



[ID] oa 



ho] oa 



ho] oa 



ho] oa 



Figure 60 - iPhone 3GS - EarPods - DC Offset Plots 



82 



a 

o 

o 
o 



o 
o 



LLJ 

w 
(3 






o 






un 




_L 



o —i 

Lf) O 






OLT1 

sa3uajno30# 



o 

saouajn330# 



_L 



o —i 

Lf) O 



O 

saouajn330# 



_L 



o —i 

Lf) O 



i r 



j L 



O —I 
Lf) O 



sa3uajn33Q# sa3uajn33Q# 



Figure 61 - iPhone 3GS - EarPods - Histograms 



83 



o 
o 

o 



T3 
O 

Q. 

i_ 
(C 
LU 

CO 

o 

CD 
(U 

c 
o 



o 



O 

<_> 

<D 

CO 



o 

-- «T 



o 

T3 



O 
<_) 

CO 






CM 



Ol 

n o 



m O 
cn <-> 
CD 

to 



01 

o 



CD 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 62 - iPhone 3GS - EarPods - SD Windows 



84 



'-' 



CD 
CO 

ll 



E 
< 



Q 
US 



<3 

CO 

'-■ 



o 
o 

Q 



CC 



CD 



01 tNt 



ho] apnmdwv 




TTT 



O 



fN E 



3' 



ho] oa 



ho] oa 



ho] oa 







ho] oa 



Figure 63 - iPad 2 - Built-in - DC Plots 



85 



o 

o 

o 
o 



o 

o 
■.■:■ 









O 



"1 T 



J L 



TT 



J L 



o —I 



o 

,-■-, 



"i r 



j L 



o —I 



i r 



j L 



o —I 



o un old 

sa3uajno30# 



saouajn330# 



saouajn330# 



sa3uajn33Q# sa3uajn33Q# 



Figure 64 - iPad 2 - Built-in - Histograms 



86 



"1 [1 



"i r 



"i — 1 



o 



"O -!■- 



O 
<_) 

to 



Ul 






J L 



J L 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 65 - iPad 2 - Built-in - SD Windows 



87 



'-' 



cm 

CD 

n 



E 
< 



Q 
US 



o 

CO 

CD 



o 
o 

Q 



LU 



o 



CD 



II 
Q 



O 



CD 



CD 

ll 



O 
m 

IS 

-■ I" I 

u l . 

CD O 

m ,; 



o 

n 

O 






o ooo 



ho] apnmdujv 




ho] oa 



ho] oa 



"i r 



1} 



u 



■r--^--! S 




J 



ho] oa 



ho] oa 



Figure 66 - iPad 2 - Earphones - DC Offset Plots 



88 



o 

o 

o 
o 



O 

o 
o 









O 



o 

a 

old 



sa3uajno30# 



_L 



o —I 



o 

01 



_L 



o —I 



1 T 



J L 



o —I 



saouajn330# 



saouajn330# 



sa3uajn33Q# sa3uajn33Q# 



Figure 67 - iPad 2 - Earphones - Histograms 



89 



3 

o 
<J 
a> 

a> 

o 

.c 

w 
(u 

c 
o 
.n 
a. 

LU 

CSI 
TJ 
nJ 

a. 


-^^^■-i 




o o o 
ta un -st 



01 

o 



O 
<_> 

Q> 

CO 



-—i "=f 



o 



o 
<_> 
CD 

00 



"1 II 






-. o 



J ll 



"1 II 



CO 



un O 

CM O 
0) 

CO 
o 



o 



--! ° 



-1 <=> 



J ll 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 68 - iPad 2 - Earphones - SD Windows 



90 



o 

CO 

cm 

o 

a. 



E 
< 






a 

CD 

CO 
CM 
CM 
O 



O 
O 

a 



ce 



LU 



o 



a 



o 



o 



o o o o 

CN [NT 



ho] spnmduuv 











_1_L 



a 

■ 

r.-i 
CN 

CT. 
l£ 
CN 

II 
Q 

m 



a 

CO 

- I 
CO 
'T. 
■.' 

CD 
II 



O 

en 

CO 



a 



r«ccaca>j 



ho] oa 




ni i 



ho] oa 



ho] oa 



CM E 



[10] oa 



Figure 69 - iPad 2 - EarPods - DC Offset Plots 



91 



o 

o 
o 



O 

o 
o 
o 
on 

T 



LU 






O 



o 

a 

old 



sa3uajno30# 



J L 



"i r 



J L 



o —I 



o 

,-■-, 



"i r 



j L 



o —I 



i r 



j L 



o —I 



saouajn330# 



saouajn330# 



sa3uajn33Q# sa3uajn33Q# 



Figure 70 - iPad 2 - EarPods - Histograms 



92 



o 
o 

o 



T3 
O 

Q. 

i_ 
(0 
LU 

CSI 
"O 

ns 
a. 



o 



o 
<_> 

to 



"i r 



J_L 



i ii i i r 



i ii i i i 



CO 

o 



o 

CO 

■ 

o 



1 — I' TT 'lrM 



I I I 



o 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 71 - iPad 2 - EarPods - SD Windows 



93 



o 



CO 
II 



E 
< 



Q 
CO 



o 

o 



o 
o 

Q 



O 

LD 
h- 

n 
co 
t-- 

CO 

II 

Q 

m 



O 

CO 

■=' 

CO 
CD 



o 



a 





CM 




CD 




^f 


o 


O 


a 


r- 


LD 


CM 



oooo 

CM CN*T 

h0] apnmdujv 



a 

LD 



II 
Q 

m 

_r 

a 



CO 

■:■. 



Cf 



CO 

cn 

LTI 



£ E 



O 

CD 



CN 
CD 



rj^caiccD-N 



[ID] DO 




^i cm 



ho] oa 



ho] oa 



[ 



CM 



Li 



1 



ho] oa 



Figure 72 - iPad 1 - Built-in - DC Offset Plots 



94 



o 

o 

o 
o 
on 

01 
01 



O 
o 

01 

01 

m 






O 






o 
a 



sa3uajno30# 



_L 



o —I 






_L 



o —I 



,-■-, 



_L 



o —I 



1 T 



J L 



o —I 



saouajn330# 



saouajn330# 



sa3uajn33Q# sa3uajn33Q# 



Figure 73 - iPad 2 - Built-in - Histograms 



95 





i -a 

^^H 




o 
o 
a> 

a> 
o 

'5 
CO 

(0 


— -j^S 


CL 


^^^fe 






^3^ 



o 



^~ o 
<_> 

(Is 

to 

LP: 



\- o 



L0 

o 

T3 



O 
<_) 

to 



I- O 



P 1 



o 



»i O 
u 

to 

I 

o 






o 



1- ° 



n 


o 


o 


D 


D 


o 


O 


o 


o 


o 


o 


o 


i.'j 


I-- 


LO 


CO 


h~ 


LQ 


CO 


h~ 


jCi 


CO 


r-~ 


LO 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 74 - iPad 2 - Built-in - SD Windows 



96 



o 

o 

■; 
:■: 

ta 



E 
< 






o 
<tm 

CM 
CO 
CD 



O 
O 
Q 




ho] apnmdwv 



ho] oa 



[ID] OQ 



[ID] 0Q 



[id] oa 



Figure 75 - iPad 2 - Earphones - DC Offset Plots 



97 



o 

o 

o 
o 



O 

o 
o 

rn 
■:■:■ 

CM 



;'l 



o 






_L 



o —I 



o 



_L 



o —I 



o 

,-■-, 



_L 



o —I 



"1 T 



J L 



o —I 



o un old 

sa3uajno30# 



saouajn330# 



saouajn330# 



sa3uajn33Q# sa3uajn33Q# 



Figure 76 - iPad 2 - Earphones - Histograms 



98 



• rri 



m 


01 




t > 




i > 




a 




<: 




"O 




i_ 




; > 




u 




id 




U) 




i~> 


1JJ 

■- 1 


oo 



+ s 



111 

o o o 

en ■* r^ 



o 



T ° 



ho] as 



ho] as 



[id] as 



[id] as 



Figure 77 - iPad 2 - Earphones - SD Windows 



99 



o 

CD 

CO 

■■: 
o 

CM 

cri 
un 

ll 



E 
< 



Q 



o 

CO 
CM 

un 

ID 
CD 
O 



O 
O 
Q 



LU 




o 


o 


CD 




cn 






L/J 




LD 



ho] apnmduuv 



ho] oa 



ho] oa 



ho] oa 



i i -n II Si 







J_L 



-1(N £ 



ho] oa 



Figure 78 - iPad 2 - EarPods - DC Offset Plots 



100 



o 

o 
o 



O 

o 
o 
o 

■" I 



LU 






O 



o 

o 

old 



"1 T 



J L 



o —I 



o 



"i r 



J L 



o —I 



o 

,-■-, 



"i r 



j L 



o —I 



i r 



j L 



o —I 



sa3uajno30# 



saouajn330# 



saouajn330# 



sa3uajn33Q# sa3uajn33Q# 



Figure 79 - iPad 2 - EarPods - Histograms 



101 





_^^v 








iw*^* 




^^^^^ 




























t ^^^ 




^ 








c 
T3 


1 — 


O 

o 

ct: 


^^Pl 


o 

? 
J? 










U 
Q. 

1— 

LU 
(0 


_.^^C__ 











"^t 











—3 1 i 








^^^^ 








^^— i 



L0 

O 



O 
<_> 

00 



^^^- J 

_^^K _ j 



o 

00 *= 



o 
<_> 

00 









I 






o 



o 
(_) 
a> 
00 
o 



o 



tij -5f 






ho] as 



ho] as 



[id] as 



[id] as 



Figure 80 - iPad 2 - EarPods - SD Windows 



102 



REFERENCES 



1. National Communications Standard. (1996) Federal Standard 1037C: Bias. 
Retrieved April 2013, from http://www.its.bldrdoc.gov/fs-1037/dir- 
004/_0587.htm 

2. Koenig, Bruce E. et al. (2012) Evaluation of the Average DC Offset Values for Nine 
Small Digital Audio Recorders. AES International Conference. Denver. 

3. Koenig, Bruce E. and Lacey, Doug S. (2013). The Average DC Offset Values for 
Small Digital Audio Recorders in an Acoustically-Consistent Environment. 

4. Fuller, Daniel B. (2012) How Audio Compression Algorithms Affect DC Offset in 
Audio Recordings. AES International Conference. Denver. 

5. Koenig, Bruce E. et al. (2012) Evaluation of the Average DC Offset Values for Nine 
Small Digital Audio Recorders. AES International Conference. Denver. 2-9. 

6. Fuller, Daniel B. (2012) How Audio Compression Algorithms Affect DC Offset in 
Audio Recordings. AES International Conference. Denver. 

7. Koenig, Bruce E. and Lacey, Doug S. (2013). The Average DC Offset Values for 
Small Digital Audio Recorders in an Acoustically-Consistent Environment. 

8. Koenig, Bruce E. and Lacey, Doug S. (2013). The Average DC Offset Values for 
Small Digital Audio Recorders in an Acoustically-Consistent Environment. 4-5. 

9. Koenig, Bruce E. et al. (2012) Evaluation of the Average DC Offset Values for Nine 
Small Digital Audio Recorders. AES International Conference. Denver. 5. 

10. Koenig, Bruce E. and Lacey, Doug S. (2013). The Average DC Offset Values for 
Small Digital Audio Recorders in an Acoustically-Consistent Environment. 4. 

11. Fuller, Daniel B. (2012) How Audio Compression Algorithms Affect DC Offset in 
Audio Recordings. AES International Conference. Denver. 

12. Koenig, Bruce E. et al. (2012) Evaluation of the Average DC Offset Values for Nine 
Small Digital Audio Recorders. AES International Conference. Denver. 5. 



103 



13. Koenig, Bruce E. et al. (2012) Evaluation of the Average DC Offset Values for Nine 
Small Digital Audio Recorders. AES International Conference. Denver. 5. 

14. Koenig, Bruce E. and Lacey, Doug S. (2013). The Average DC Offset Values for 
Small Digital Audio Recorders in an Acoustically-Consistent Environment. 4. 

15. Koenig, Bruce E. et al. (2012) Evaluation of the Average DC Offset Values for Nine 
Small Digital Audio Recorders. AES International Conference. Denver. 5. 

16. Fuller, Daniel B. (2012) How Audio Compression Algorithms Affect DC Offset in 
Audio Recordings. AES International Conference. Denver. 

17. Koenig, Bruce E. and Lacey, Doug S. (2013). The Average DC Offset Values for 
Small Digital Audio Recorders in an Acoustically-Consistent Environment. 

18. Koenig, Bruce E. et al. (2012) Evaluation of the Average DC Offset Values for Nine 
Small Digital Audio Recorders. AES International Conference. Denver. 5. 

19. Grigoras, Catalin. Rappaport, Daniel. Smith, Jeff M. (2012) Analytical Framework 
for Digital Audio Authentication. AES International Conference. Denver. 

20. Koenig, Bruce E. et al. (2012) Evaluation of the Average DC Offset Values for Nine 
Small Digital Audio Recorders. AES International Conference. Denver. 9. 



104